A survey on human face recognition invariant to illumination


Published on

Published in: Technology, News & Politics
  • Be the first to comment

  • Be the first to like this

No Downloads
Total views
On SlideShare
From Embeds
Number of Embeds
Embeds 0
No embeds

No notes for slide

A survey on human face recognition invariant to illumination

  1. 1. International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 2, March – April (2013), © IAEME517A SURVEY ON HUMAN FACE RECOGNITION INVARIANT TOILLUMINATIONU.K. Jaliya1, J.M. Rathod21Assistant Professor, Department of Computer Engineering, BVM Engineering College,Vallabh Vidyanagr, Anand, Gujarat, India2Associate Professor, Department of Electronics, BVM Engineering College, VallabhVidyanagr, Anand, Gujarat, IndiaABSTRACTHuman face recognition is one of the research areas in the current era of the research.It is one the widely used biometric technique for identification and verification of the humanface. There are many challenges to face recognition which degrade the performance of thealgorithm. The illumination variation problem is one of the well-known problems in facerecognition in uncontrolled environment. In this paper an extensive and up-to-date survey ofthe existing techniques to address this problem is presented. Different authors have given somany techniques for illumination reduction from the face image but still some combinedsurvey is missing so we have tried to fill that gaps in this paper. We have collected variouspreprocessing techniques suggested by different authors and shown their results in a tabularform. After preprocessing we can use any of the recognition method for face recognition.There are so many online face databases available so we can use any of them.Keywords: PCA (Principle component Analysis), HE (Histogram Equalization), AHE(Adaptive Histogram Equalization), BHE (Block-based Histogram Equalization), DWT(Discrete Wavelet Transform), DCT (Discrete Cosine Transform), LBP (Local BinaryPattern), DMQI (Dynamic Morphological Quotient Image), LTP (Local Ternary Pattern),DSFQI (Different Smoothing filters Quotient Image).I. INTRODUCTIONFace recognition has been an active research area over the last 30 years. It has beenstudied by scientists from different areas of psychophysical sciences and those from differentareas of computer sciences. Psychologists and neuroscientists mainly deal with the humanINTERNATIONAL JOURNAL OF COMPUTER ENGINEERING& TECHNOLOGY (IJCET)ISSN 0976 – 6367(Print)ISSN 0976 – 6375(Online)Volume 4, Issue 2, March – April (2013), pp. 517-525© IAEME: www.iaeme.com/ijcet.aspJournal Impact Factor (2013): 6.1302 (Calculated by GISI)www.jifactor.comIJCET© I A E M E
  2. 2. International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 2, March – April (2013), © IAEME518perception part of the topic, whereas engineers studying on machine recognition of humanfaces deal with the computational aspects of face recognition.Face recognition has applications mainly in the fields of biometrics, access control,law enforcement, and security and surveillance systems. Biometrics are methods toautomatically verify or identify individuals using their physiological or behavioralcharacteristics [1].The necessity for personal identification in the fields of private and secure systemsmade face recognition one of the main fields among other biometric technologies. Theimportance of face recognition rises from the fact that a face recognition system does notrequire the cooperation of the individual while the other systems need such cooperation.Figure 1.shows the sketch for any pattern recognition technique.Figure 1 Sketch of pattern recognition architectureThe topic seems to be easy for a human, where limited memory can be a mainproblem; whereas the problems in machine recognition are manifold. Some of possibleproblems for a machine face recognition system are mainly;1) Facial expression change2) Illumination change3) Aging4) Rotation:5) Size of the image6) Frontal vs. ProfileII. ILLUMINATION PROCESSIONGIllumination variation is one the main challenging problem in any face recognitionsystem. There are two main approaches for illumination processing: Active approach andPassive approach [2].Active approaches apply active sensing techniques to capture faceimages which are invariant to environment illumination. Passive approach attempt toovercome illumination variation in face images due to environment illumination change andwe will focus on this approach. Figure 2 is the framework of any face recognition methodsinvariant to illumination.
  3. 3. International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 2, March – April (2013), © IAEME519Figure 2. Framework of face recognition methodsIII. PRE-PROCESSINGOften images are not in the correct form to be plugged straight into a recognitionalgorithm. The image may contain more information than one single face, and the lightingconditions in the test image may not be the same as in the sample data for training thealgorithm. This can greatly affect the effectiveness or the recognition rate of the algorithm.Therefore, to obtain the best possible results, it is necessary to pre-process an image tonormalize lighting and remove noise before inserting it into a recognition algorithm. Manyresearchers have used so many techniques some of which are explained in following section.1) Histogram Equalization (HE)Histogram equalization (HE) is a classic method. It is commonly used to make animage with a uniform histogram, which is considered to produce an optimal global contrast inthe image. However, HE may make an image under uneven illumination turn to be moreuneven [3].Figure 3. Histogram equalization of an image2) Adaptive Histogram Equalization (AHE)It computes the histogram of a local image region centered at a given pixel todetermine the mapped value for that pixel; this can achieve a local contrast enhancement.However, the enhancement often leads to noise amplification in “flat” regions, and “ring”artifacts at strong edges. In addition, this technique is computationally intensive [4].
  4. 4. International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 2, March – April (2013), © IAEME5203) Block-based Histogram Equalization (BHE)This method is also called local histogram equalization or region based histogramequalization. The face image can be divided into several small blocks according to thepositions of eyebrows, eyes, nose and mouth. Each block is processed by HE. In order toavoid the discontinuity between adjacent blocks, they are overlapped by half with each other.BHE is simple so that the computation required of BHE is much lower than that of AHE. Thenoise produced by BHE is also very little.4) LogAbout MethodThe LogAbout method which is an improved logarithmic transformations as thefollowing equation:cbyxfayxgln)1),(ln(),(++= (1)Where g(x, y) is the output image; f(x, y) is the input image; a, b and c are parameters whichcontrol the location and shape of the logarithmic distribution. Logarithmic transformationsenhance low gray levels and compress the high ones. They are useful for non-uniformillumination distribution and shadowed images. However, they are not effective for highbright images [6]. An example of what the LogAbout algorithm does can be seen in belowfigure 4.Figure 4. LogAbout Illumination Normalization5) Sub-Image Homomorphic FilteringIn Sub-Image Homomorphic filtering method, the original image is split vertically intwo halves, generating two sub-images from the original one (see the upper part of belowfigure 5). Afterwards, a Homomorphic Filtering is applied in each sub-image and theresultant sub-images are combined to form the whole image. The filtering is subject to theillumination reflectance model as follows:),(),(),( yxLyxRyxI = (2)Where I(x, y) is the intensity of the image; R(x, y) is the reflectance function, which isthe intrinsic property of the face; L(x, y) is the luminance function. Based on the assumptionthat the illumination varies slowly across different locations of the image and the localreflectance changes quickly across different locations, a high-pass filtering can be performedon the logarithm of the image I(x, y) to reduce the luminance part, which is the low frequencycomponent of the image, and amplify the reflectance part, which corresponds to the highfrequency component [7].Similarly, the original image can also be divided horizontally (see the lower part ofbelow figure 5) and the same procedure is applied. But the high pass filter can be different.At last, the two resultant images are grouped together in order to obtain the output image.
  5. 5. International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 2, March – April (2013), © IAEME521Figure 5. Sub-Image Homomorphic Filtering6) Discrete Cosine Transform (DCT)In a face image, illumination usually changes slowly compared with the reflectanceexcept some casting shadows and secularities on the face. As a result, illumination variationsmainly lie in the low-frequency band. Therefore, we can remove the low frequency part toreduce illumination variation. The low frequency DCT coefficients are set to zero toeliminate illumination variations. Figure 6 shows the images with various illuminationconditions and normalized images using DCT components [8].Figure 6. Original and DCT applied images7) Discrete Wavelet Transform (DWT)Besides the DCT, discrete wavelet transform (DWT) is another common method inface recognition. There are several similarities between the DCT and the DWT: 1) They bothtransform the data into frequency domain; 2) As data dimension reduction methods, they areboth independent of training data compared to the PCA. Because of these similarities, thereare also several studies on illumination invariant recognition based on the DWT. Similar tothe idea in (Chen et al., 2006), a method on discarding low frequency coefficients of theDWT instead of the DCT was proposed (Nie et al., 2008). Face images are transformed fromspatial domain to logarithm domain and 2-dimension wavelet transform is calculated by the
  6. 6. International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 2, March – April (2013), © IAEME522algorithm. Then coefficients of low-low sub band image in n-th wavelet decomposition arediscarded for face illumination compensation in logarithm domain. The experimental resultsprove that the proposed method outperforms the DCT and the quotient images. The kind ofwavelet function and how many levels of the DWT need to carry out are the key factors forthe performance of the method. Different from the method in (Nie et al., 2008), Han et al.(2008) proposed that the coefficients in low-high, high-low and high-high sub band imageswere also contributed to the effect of illumination variation besides the low-low sub bandimages in n-th level. Based on the assumption, a homomorphic filtering is applied to separatethe illumination component from the low-high, high-low, and high-high sub band images inall scale levels. A high-pass Butterworth filter is used as the homomorphic filter. Figure 7shows the wavelet decomposition of an image.Figure 7. Wavelet decomposition of an image.The novelty of the DWT method is that the light variation in an image can be modeledas multiplicative noise and additive noise, instead of only the multiplicative term in whichmay be instructive in modeling the face under illumination variations in future.However, by comparing the results of the DWT method and the DCT, we find the result ofthe DWT method is worse than that of the DCT. Hence, the DWT method is not as effectiveas the DCT for illumination invariant recognition.8) Dynamic Morphological Quotient Image (DMQI) methodIn the framework of Lambertian model, the intensity of point (x, y) in an image I ismodeled as shown in equation (1), where R and L are components of reflectance andillumination respectively. Estimating R and L only from the input image I is a well-known ill-posed problem. Therefore there are two assumptions: (1) the illumination L is smooth and (2)the reflectance R can be varied randomly [9].According to the Lambertian reflectance theory, R depends on the albedo and surface normaland hence is the intrinsic representation of an object. It is R that represents the identity of aface. L is the illumination and is the extrinsic factor.Dynamic Morphological Quotient Image (DMQI) method in which mathematicalmorphology operation is employed to smooth the original image to obtain a better luminanceestimate. However, in DMQI, there is some pepper noise in dark area.
  7. 7. International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 2, March – April (2013), © IAEME5239) Different Smoothing filters Quotient Image (DSFQI)Different Smoothing filters Quotient Image (DSFQI) give a new demonstration inwhich it is proved that the same smoothing functions with different scales also contribute toextract the illumination invariant features. Figure 8 denotes this merit. It tends to retain theremarkable face features such as edges and verges, and gains the intrinsic representation of anobject [10].Figure 8. Original, DMQI and DSFQI images10) Local Binary Pattern (LBP)Local Binary Pattern (LBP) (Ojala et al., 2002) is a local feature which characterizesthe intensity relationship between a pixel and its neighbors. The face image can be dividedinto some small facets from which LBP features can be extracted. These features areconcatenated into a single feature histogram efficiently representing the face image. LBP isunaffected by any monotonic grayscale transformation in that the pixel intensity order is notchanged after such a transformation [11].11) Improved Local Binary Pattern (ILBP)In order to utilize the excellent discriminative power and computational simplicity ofthe LBP descriptor, while abating the performance degradation due to varying illuminationconditions for face recognition. We have shown an enhanced LBP-based face recognitionalgorithm that fuses illumination invariant DMQI with discriminative LBP descriptors. In ourDMQI-LBP algorithm, illumination variations in face images are first normalized with theDMQI, then the DMQI is segmented into 7×7 sub-blocks, and uniform patterns are extractedin these sub-blocks to form the LBP feature histograms. Finally, LBP histograms from allsub-blocks are concatenated into face feature vectors, and a weighted chi-square distancemetric which considers the different roles of each sub-block for face recognition, is evaluatedto measure the similarity between a probe face feature and the stored subject face features[11].12) Local Ternary Pattern (LTP)A local ternary pattern (LTP), another important extension of original LBP isproposed. The most important difference between the LTP and LBP is that the LTP use 3-valued codes instead 2-valued codes in the LBP. Because of the extension, the LTP is morediscriminant and less sensitive to noise. To apply the uniform pattern in the LTP, a codingscheme that split each ternary pattern into its positive and negative halves is also proposed in(Tan & Triggs, 2010). The resulted halves can be treated as two separated LBPs and used forfurther recognition task. The local directional pattern is more robust against noise and non-monotonic illumination changes [11].
  8. 8. International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 2, March – April (2013), © IAEME524IV. COMPARISON OF DIFFERENT PRE-PROCESSING METHODSHere in this section we are comparing the performance results given by variousauthors in their research paper. To evaluate the performances of different methods undervarying lighting conditions without other variances, there are four popular face databases, theYale B, Extended Yale B, CMU PIE and FERET database. In the Yale Face database B, thereare 64 different illumination conditions for nine poses per person. To study the performancesof methods under different light directions, the images are divided into 5 subsets based on theangle between the lighting direction and the camera axis. The Extended Yale B databaseconsist 16128 images of 28 subjects with the same condition as the Original Yale B. In theCMU PIE, there are altogether 68 subjects with pose, illumination and expression variations.The performances of several pre-processing techniques are shown in table 1.This table showsthe error rate calculated by different methods by varying images in the subset from differentdatabases. The dotted lines in the table in one row indicate the result obtained by extendedYale B database. In table ‘n/a’ indicate that author has not used that database for performancemeasure. In method column in table ‘+‘is used for combining preprocessing methods.Table 1 Performance Comparisons of Different MethodsV. CONCLUSIONThe modeling approach is the fundamental way to handle illumination variations, butit always takes heavy computational burden and high requirement for the number of trainingsamples. The LBP is an attractive area which can tackle illumination variation coupled withother variations such as pose and expression. For normalization methods, the methods ondiscarding low-frequency coefficients are simple but effective way to solve the illuminationvariation problem. However, a more accurate model needs to be studied instead of simplydiscarding low-frequency coefficients. However, each technique still has its own drawbacks.MethodsError Rate (%)Yale B / Yale B + Extended Yale B CMUPIEFERETSub set3 Sub set4 Sub set5Non 10.8 51.4 77.4 43.0 n/aRHE 17.8 71.1 79.4 14.6 n/aLogAbout 14.4 42 30.7 43.0 n/aLOG+DCTn/a n/a n/a0 n/a12.9 12.4 15.2HE9.2 54.2 41.147.8 n/a62.3 78.7 89.9DCT0 0.18 1.71 0.36n/a16.4 14.5 16.2DCT+LBP 10.12 15.33 17.29 n/a n/aDWT 0 0.19 0.53 n/a n/aLTVn/a n/a n/a0n/a20.6 23.9 21.7DCT+PCA n/a 5.84DMQI 2.98 3.98 4.93 n/a n/aDSFQI 1.22 1.62 1.76 n/a n/a
  9. 9. International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 2, March – April (2013), © IAEME525VI. ACKNOWLEDGEMENTSWe are very thankful to our principal Dr. F.S. Umrigar and Prof.P.B.Swadas, HeadComputer Engineering Department, BVM Engineering College for encourages us to writethis review paper.REFERENCES[1]. P. S. Huang, C. J. Harris, and M. S. Nixon, “Human Gait Recognition in CanonicalSpace using Temporal Templates”, IEE Proc.-Vis. Image Signal Process, Vol. 146, No.2, April 1999.[2]. Gonzales, R.C. & Woods, R.E. (1992). Digital Image Processing, second ed., PrenticeHall, ISBN-10: 0201180758, Upper Saddle River.[3]. Pizer, S.M. & Amburn, E.P. (1990). Adaptive Histogram Equalization and ItsVariations. Comput. Vision Graphics, Image Process, 39, pp. 355-368.[4]. Xie, X. & Lam, K.M. (2005). Face Recognition under Varying Illumination based on a2D Face Shape Model, Pattern Recognition, 38(2), pp. 221-230.[5]. Liu, H.; Gao, W.; Miao, J.; Zhao, D.; Deng, G. & Li, J. (2001).IlluminationCompensation and Feedback of Illumination Feature in Face Detection, Proc. IEEEInternational Conferences on Info-tech and Info-net, 23, pp. 444-449.[6]. Delac, K.; Grgic, M. & Kos, T. (2006). Sub-Image Homomorphic Filtering Techniquesfor Improving Facial Identification under Difficult Illumination Conditions.International Conference on System, Signals and Image Processing, Budapest.[7]. Shermina.J, (2011). Illumination Invariant Face Recognition Using Discrete CosineTransform And Principal Component Analysis.IEEE Proceedings Of Icetect.[8]. X.G. He, J. Tian, L.F. Wu, Y.Y. Zhang, X. Yang l. llumaination Normalization withMorphological Quotient Image,” Journal ofSoftware, Vol.18(9) pp.2318-2325, 2007.[9]. Yu CHENG, Zhigang JIN, Cunming HAO (2010)“Illumination Normalization based onDifferent Smoothing Filters Quotient Image” IEEE Third International Conference onIntelligent Networks and Intelligent Systems.pp.28-31.[10]. PAN Hong, XIA Si-Yu, JIN Li-Zuo, XIA Liang-Zheng School of Automation,Southeast University, Nanjing 210096, P. R. China “Illumination Invariant FaceRecognition Based on Improved Local Binary Pattern” Proceedings of the 30th ChineseControl IEEE Conference Jul y 22-24, 2011, Yantai, China.[11]. K. Ramachandra Murthy, Asish Ghosh “An Efficient Illumination Invariant FaceRecognition Technique using Two DimensionalLinear Discriminant Analysis” 1st Int’lIEEE Conf. on Recent Advances in Information Technology | RAIT-2012 |.[12]. Prof. B.S Patil and Prof. A.R Yardi, “Real Time Face Recognition System using EigenFaces”, International Jjournal of Electronics and Communication Engineering &Technology (IJECET), Volume 4, Issue 2, 2013, pp. 72 - 79, ISSN Print: 0976- 6464,ISSN Online: 0976 –6472.[13]. Steven Lawrence Fernandes and Dr. G Josemin Bala, “Analysing Recognition Rate ofLDA and LPP Based Algorithms for Face Recognition”, International Journal ofComputer Engineering & Technology (IJCET), Volume 3, Issue 2, 2012, pp. 115 - 125,ISSN Print: 0976 – 6367, ISSN Online: 0976 – 6375.